Cloudflare Workers MCP — Edge Functions for AI Agents

What it is

Cloudflare Workers MCP is a Model Context Protocol server that connects AI coding agents to the Cloudflare developer platform. It lets agents deploy edge functions, manage KV key-value storage, R2 object storage, and D1 SQLite databases through natural language commands.

This tool is designed for developers already using Cloudflare who want AI-assisted serverless deployment. It works with Claude Code, Cursor, and any MCP-compatible client. Setup takes under two minutes.

How it saves time or tokens

Without this MCP server, deploying a Cloudflare Worker requires switching between the Wrangler CLI, the Cloudflare dashboard, and your code editor. The MCP server collapses that workflow into a single conversation. Your agent writes the function, deploys it, and configures storage in one pass. The workflow estimates around 2,400 tokens per session, covering deployment, storage configuration, and verification steps.

How to use

1. Get an API token from dash.cloudflare.com/profile/api-tokens with Workers and Storage permissions.
2. Add the MCP server to your .mcp.json configuration file.
3. Restart your MCP client and start issuing natural language commands.

{
  "mcpServers": {
    "cloudflare": {
      "command": "npx",
      "args": ["-y", "@cloudflare/mcp-server-cloudflare"],
      "env": {
        "CLOUDFLARE_API_TOKEN": "your-token"
      }
    }
  }
}

Example

Once configured, you can issue commands like these to your AI agent:

'Deploy a new worker that returns the current time in JSON'
'Store user preferences in KV namespace settings'
'List all objects in the backups R2 bucket'
'Create a D1 database called analytics and add a pageviews table'

The agent translates these into Cloudflare API calls, executes them, and reports results back in the conversation.

Related on TokRepo

• MCP Integrations — Browse other MCP server integrations for AI agents
• AI Tools for DevOps — More developer operations tools for AI-assisted workflows

Common pitfalls

• API tokens need explicit permissions for each service (Workers, KV, R2, D1). A token missing R2 permissions will silently fail on bucket operations.
• The MCP server runs npx on every invocation, which adds cold-start latency. Pin the package version to avoid unexpected updates.
• D1 databases have a 500MB size limit on the free plan. Large datasets should use R2 or external storage instead.